In an intensive meeting in November 1984, the ASML team plotted its product roadmap in a day. The actions and results that followed formed the basis of the time-to-market attitude and culture that still exists in the company today.
It happened in just one day and it must have been a weird spectacle. Unfortunately, I never got to speak to the two Philips trainers, both outsiders, who witnessed what can be described as the coming of age of ASML. None of the interviewees for the ASML book remembered who these two guys were, but in mid-November 1984, they were present when all the pieces fell together after half a year of preparation, writing plans, doubling the ASML staff and getting sufficient financial resources. This is what I heard from a handful of ASML people who were there as well.
In mid-November 1984, ASML’s managers and senior engineers boarded a bus headed north. The twenty of them were going to spend the next two days in Hotel Victoria at the edge of De Hoge Veluwe National Park, engaged in a mix of team building and training. They were preparing for an intense bout of catch-up – development had to move faster. Their new machine had to be ready for launch in record time.
Their goal was clear. Philips Elcoma wanted to receive the PAS 2500 stepper on 1 April 1986, and a few months later, ASML wanted to demo the machine at Semicon West. That gave them a year to design and manufacture a new system capable of handling the generation of chips that fabs would be starting to test in 1986. ASML’s managing director Gjalt Smit had given them a deadline of 1 January 1986. In the park-side hotel, the company’s leadership discussed how to make that happen.
The next step
This tech-savvy crew had already picked the technology to do the trick. They were going to use the i-line wavelength from a mercury bulb to get to a resolution of less than a micrometer. It was the next step from the mercury g-line the chip industry used at the beginning of the 80s.
Around 2:00 pm the same day, the team building program started. Two trainers, both from Philips, kicked off with a standard course on how to move from idea to completed machine faster. At Philips, that process typically took nine to twelve years, and the company had come up with a method to reduce that to roughly seven years. If ASML was to meet the Elcoma deadline, it only had a little over one year.
Just when they were ready to continue with their second day, Smit stepped forward, announcing the training was over. Last evening, he had returned to Schiphol airport, after visiting US customers. He had some news: there was no time to lose. The whole team was ordered to rethink the product roadmap immediately.
A machine no one will buy
Smit told them rather bluntly that if ASML would keep pursuing the i-line stepper, it might be the technological frontrunner, but it would have a machine no one would buy. The whole industry had set up its infrastructure and available materials based on the g-line.
Smit had another announcement. During his trip to the US, he’d become convinced that ASML really couldn’t sell the current PAS 2000 stepper as a test machine. The oil controlling the movements was simply a no-go in fabs.
Smit’s conclusion: they needed a machine without oil, preferably the following year. Customers had again pointed out ASML’s lack of an installed base, and so he wanted a machine they could start experimenting with as soon as possible. Waiting for the development of the PAS 2500 to complete wasn’t an option.
To make the step to machines that were ready for sale, Smit wanted a stepper to demo at Semicon West 1985. Everybody thought: how on earth are we going to complete a machine in just over six months? Smit was asking the all-but-impossible.
In the pursuing discussion, Smit refused to budge on the deadline. He wanted to demonstrate a g-line stepper at Semicon West the next spring, period. Without compromising the development on the PAS 2500 – because after a first machine in 1985, ASML had to push forward with a system for the VLSI generation of chips. “How are we going to have a working g-line machine seven months from now?,” he asked. “I want a plan by the end of the day.”
Otherwise, we’ll just go bankrupt
Then freshman, now CTO, Martin van den Brink was also present. Still impressed by the dynamics of his boss at the time, Van den Brink said in August 2015: “Gjalt wanted us to speed up. He had just arrived by plane and I thought: why is he doing this? In retrospect, that was an important moment. We were sitting with colleagues who had never worked with each other, none of us had the slightest idea of steppers. I remember that like it was yesterday.”
Wim Hendriksen, a software guy, was also there: “Gjalt said: customers want us to deliver earlier. As down-to-earth Dutchmen, we had nothing to do with customers, but Gjalt did. So he entered the training and turned everything upside down. He said: I shall have such a machine before May 1985, otherwise, we’ll just go bankrupt. Well, at the end of the day, it was a gigantic mess. When we left the building, the paper was stacked a meter high. But we had a viable plan.”
The plan came down to exchanging the oil-driven stage in the existing PAS 2000 stepper with an electric one. That technology existed but hadn’t been used by ASML yet. There was only a proof of concept for these linear motors at the Philips Research Laboratories. It still had to be productized. ASML eventually did succeed in developing this intermediate machine PAS 2400. It was basically a PAS 2000 stepper with a new wafer stage controlled by linear motors.
Two months after Semicon West 1985, three chip manufacturers showed interest in the PAS 2400s: AMD, Cypress and National Semiconductor. In October 1985, PAS 2400s were running at Philips fabs in Nijmegen and Hamburg and Philips Research in Eindhoven. ASML even was chosen as one of four litho companies to deliver a system for next-generation technology development at AMD.
The Hotel Victoria adventure and the resulting actions and success proved an important lesson for the ASML team. To serve and satisfy customers, you have to get rid of the attitude that it has to be 100 percent perfect all the time. In fact, sometimes a sloppy machine is no problem, as long as it is in time and works sufficiently well. This applies to startups in general but to litho in particular.
After its startup years, ASML managed to beat the competition by getting the right technology to customers in time, not by making the most reliable machines. This has been the case with each generation of steppers and scanners that followed the PAS2400, up to EUV. You might argue the game changed somewhat during the evolution of the current extreme generation. To be recognized as EUV users, customers were even prepared to receive systems that didn’t work at all for years. It was more important to show off and send a message to their customers, saying: we’re committed to continuing on the path of Moore’s Law.